On 14 May 2012, at 16:10, Camillo Bruni wrote:
maybe we should even start considering again replacing dict + set with something more standard that internally relies on buckets instead of associations.
I've shown in my master thesis that the bucket implementation performs slightly better or the at least the same in various use-cases.
Better implementations would always be welcome. For example, choosing between regular and small dictionary should be an automatic internal detail. Exposing associations as an implementation details is of course wrong. But one could keep the association concept at the API level and do a more intelligent implementation. But that doesn't change the value of the original issue: Dictionary>>#add: and Dictionary>>#addAll: do conflict in a certain sense. Sven